Philia

Inspiration

Imagine you've just moved to a new city for an internship, new join, new area. It's 6 PM on a Tuesday. You're hungry, you don't know anyone, and the thought of eating dinner alone in your apartment for the fifth night in a row feels heavier than it should.

That feeling isn't rare — it's a public health crisis. In 2023, the U.S. Surgeon General declared loneliness a public health epidemic. One in two young adults report feeling lonely. And the cure isn't another app to scroll through — it's the opposite. It's in-person connection, and the research is unambiguous: even one meaningful face-to-face interaction a week measurably improves long-term wellbeing.

But every existing app pushes us in the wrong direction. Dating apps optimize for romance. Social platforms optimize for screen time. Event apps require planning days in advance and often feel like networking. Nothing exists for the simplest, most human use case: I want to share a meal with someone, tonight.

We named the app Philia (φιλία) — the Greek word for the love between friends, the warmth of companionship. Aristotle considered it the highest form of social bond. We wanted to build the version of social technology that earns that name: one that gets you off your phone and across a table from someone new.

What it does

Philia is a mobile app for same-day, spontaneous dinners with compatible strangers nearby.

Think Beli x Yelp x Snapchat.
Open the app at 6 PM, see every dinner plan happening in your area in the next two hours.
Join a group in one tap — our matching algorithm scores compatibility on food preferences, budget, group size, and vibe, so the table is blind, but compatible.
Or start your own — pick the restaurant, the time, the price range, and the group size, and Philia surfaces it to people whose preferences line up with yours.
No DMs, no swiping. A built-in chatbot runs the entire group chat as a neutral coordinator — confirming the table, nudging flakers, sending meeting instructions, doing mood checks before the meal — so the actual conversation happens face-to-face, not on a screen.
A live map shows active groups across the city, with custom pins displaying the restaurant and how many seats are left.
A reliability score on every profile tracks attendance vs. flakes, keeping the community accountable and welcoming.
Small groups only — gatherings are capped at 6 to stay intimate and low-pressure, especially for introverts.

The whole experience is designed so a user can go from opening the app to sitting at a confirmed dinner in under 30 seconds. The idea is to shift the mental load of planning in advance, to spontaneous SAME DAY schedling.

How we built it

Frontend: React Native with Expo for cross-platform mobile, styled in a "Notion-inspired" minimalist aesthetic — soft pastel palette, heavily rounded cards, custom pill-shaped map pins. [VERIFY stack]
Backend: Firebase for auth, real-time database for active groups, and cloud functions for the matching pipeline. [VERIFY]
Maps: Mapbox for the discovery view, with custom overlays rendering live groups as interactive pins. [VERIFY]
Matching algorithm: A weighted compatibility scorer that takes food preferences, dietary restrictions, budget range, group size, and vibe tags, and ranks candidate groups against a user's profile. We deliberately chose a deterministic, transparent scoring system over a black-box approach — when you're putting strangers at a table together, predictability and explainability matter more than novelty.
Chatbot coordinator: A scripted conversation flow that mediates the entire group experience — collecting RSVPs, confirming the table, sending logistics, nudging quiet members, and pinging anyone who's late. The chatbot has a consistent friendly tone but follows a structured decision tree, which keeps interactions predictable and safe.
Profile system: Stylized animal avatars instead of photos to reduce shallow judgment, with a percentage-ring chart visualizing the user's attendance vs. flake history.

Challenges we ran into

Designing a chatbot that feels warm, not robotic. Our first script iterations felt like a customer service bot. We rewrote the chatbot's voice several times — adding small touches of personality, contextual phrasing, and just enough variability — to make it feel like a friend organizing dinner instead of an automated reminder system.
Balancing "blind" with "compatible." If matching is too transparent, users start optimizing their profiles like a dating app. If it's too opaque, they don't trust it. We landed on showing users why a group was suggested ("matched on cuisine + budget") without revealing other members' identities, which preserved the spontaneity while building trust.
Tuning the matching weights. Early versions over-prioritized food preferences and ended up clustering people who were demographically similar — exactly what we didn't want. We re-weighted vibe and budget higher than cuisine, which produced more diverse and surprising tables.
The flake problem. A spontaneous dinner app collapses if half the table no-shows. We debated penalty systems, deposits, and verification before landing on the public reliability score — visible attendance vs. flake count — as the lightest-weight nudge that still creates real accountability.
Real-time group state. Groups form, fill, and expire within hours. Keeping the map and the join feed accurate without aggressive polling took careful real-time database design. [VERIFY — describe what actually broke]
Scoping under hackathon time pressure. We had a much larger feature set on the whiteboard (event-style group dinners, recurring meetups, neighborhood leaderboards) and had to ruthlessly cut to land the core same-day-dinner loop in time.

Accomplishments that we're proud of

A working end-to-end loop. From signup → discovering a group on the map → joining → chatbot-mediated logistics → confirmed plan, the full flow works in a real demo.
A coordinator that genuinely changes the social dynamic. Watching the chatbot handle a group chat — confirming the table, nudging a quiet member, deflecting an awkward attempt to share contact info — and seeing it actually feel like a friend organizing dinner was the moment the product clicked.
A design language that matches the mission. The warm, minimalist, animal-avatar aesthetic genuinely lowers the social anxiety of meeting strangers. We're proud of how the visual choices reinforce the trust model.
Real engagement with the social impact thesis. We didn't bolt "fights loneliness" onto a product idea — every feature, from the small group cap to the no-DM rule to the reliability score, was designed against research on what actually creates meaningful connection.

What we learned

Predictable beats clever in stranger interactions. When users are about to meet people they don't know, what they want from technology is consistency — not surprise. Our deterministic chatbot and transparent matcher built more trust than a "smarter" system would have.
Subtraction beats addition. Every time we considered adding a feature (in-app messaging, photo sharing, ratings, social feed), it weakened the product. The strongest version of Philia is the one that pushes you off the screen the fastest.
Trust is built by visible structure, not promises. Users don't trust a "we'll keep you safe" claim. They trust seeing other users' attendance records, knowing groups are capped at 6, and knowing a chatbot — not another stranger — handles the chat. Architectural trust > rhetorical trust.
The right defaults matter more than the right options. A 2-hour spontaneous window, 4-person default group size, and same-neighborhood radius made the product feel decisive. More configurability would have made it feel like work.

What's next for Philia

Beyond dinner. The same-day, blind-but-compatible model extends naturally to coffee, workouts, study sessions, museum trips. Dinner is the wedge; co-presence is the product.
University and onboarding partnerships. New students, new interns, and people relocating for jobs are the highest-pain users. We want to partner with universities and corporate HR onboarding teams to seed Philia in the moments people most need it.
A smarter coordinator. As the product grows, we want to evolve the chatbot into something more context-aware — recognizing when a group is going quiet, suggesting a tweak to the meeting time when weather changes, sending lightweight post-dinner reflections.
A reliability layer that protects without punishing. We want to evolve the flake score into something that distinguishes between "had to bail because of a real emergency" and "ghosted." Fairness here matters as much as accountability.
Cross-cultural expansion. The loneliness epidemic isn't American. Same-day dining culture is even stronger in Tokyo, Seoul, and parts of Europe — and we want to localize Philia for cities where the cultural soil is even more fertile than the U.S.

The bigger vision: Philia is a bet that the right role for technology in social life isn't to be the friend, but to lower the activation energy of making one. Every line of code is in service of a single moment — someone closing the app, walking out the door, and sharing a meal with someone they didn't know an hour ago.